Too many words breaks it. Wrong order changes the feel. v4 and v5 respond completely differently to the same prompt — something that works great on v4.5 produces generic results on v3.
So I built Supre. You pick genre, mood, BPM, instruments, vocal style and which Suno version you're using — it generates a calibrated Style prompt specifically for that version.
Built with Claude Haiku on the backend, Vercel serverless, Upstash Redis for rate limiting.
Try it: https://supre.online/en/tool (2 free generations/day, no signup)